According to the Cramer-Rao lower bound, as SNR decreases the standard deviation in range should increase. After testing using two TREK1000 devices, the range displayed on the LCD never seemed to have a very high standard deviation even close to the point where the two devices could no longer communicate.
Why might this be?
First of all, could I ask you what software is loaded in the EVB1000s? is it version 2.10 ( as boards are straight from the box) , or would you have EVK1000 Software loaded onto the boards.
2ndly, are the boards on a height to reduce the impact on reflections and bouncing signals?
Also, could you explain what you consider as high deviation? The chip has an accuracy of a couple of centimetres at short distance apart (10cm -20 cm). See also DW10000 datasheet.
If you really want to test proximity then I suggest to use the TREK boards as EVK, by loading EVK software and calibrate them if required.